2.7 — The same words, three assistants: prompting ChatGPT, Claude & Gemini 🖥️

🟢 In plain words — you don’t have to relearn prompting for each tool. The techniques are the same everywhere; only the strengths of each assistant and the mode (fast vs reasoning) you use change.

Good news: everything you’ve just learned (the 4 ingredients, the few-shot, the CoT…) works everywhere. A good prompt is portable — it travels from ChatGPT to Claude to Gemini without rewriting. No need to learn “a dialect” per tool.

🧳 The prompt is a suitcase, not a lock. You pack a well-organized suitcase (Role, Context, Task, Format); it opens in any hotel. The differences between assistants play out at the margins — on what each does best, not on the grammar of the prompt.

(The three big ones were introduced in Level 1 — remember: product ≠ model. Here, we look at their differences when you prompt them.)

Where each one shines (mid-2026 state)

Assistant (lab)	Where it shines	Engines (mid-2026)
🟢 ChatGPT (OpenAI)	Versatility, ecosystem, multimodal (understands text, image, and voice), fast execution	GPT-5.5 (flagship) · Instant (fast) / Thinking·Pro (reasoning) variants. By default, often a fast model (e.g. GPT-5.3 Instant)
🟣 Claude (Anthropic)	Polished writing, long-form reasoning, code, instruction-following, safety	Opus 4.8 (powerful) · Sonnet 4.6 (balanced) · Haiku 4.5 (lightweight)
🔵 Gemini (Google)	Very large context, “anything-to-anything” multimodal, Google integration (Docs, Gmail)	Gemini 3.5 Flash · 3.1 Pro (reasoning) · Omni (multimodal)

📌 None is “the best” in absolute terms. For a polished email, many prefer Claude; to analyze 12 screenshots, Gemini or ChatGPT multimodal; for a draft integrated into your Google tools, Gemini. The right reflex: test 2 candidates on YOUR task, don’t trust a ranking.

The detail that changes your prompt: fast mode vs reasoning mode

Remember Rule 3 (CoT). Each assistant offers two families of engines:

• a fast mode (“Instant” on ChatGPT, Flash on Gemini, Haiku on Claude) — a standard model: explicit CoT helps (“think step by step”).
• a reasoning mode (“Thinking/Pro” on ChatGPT, Pro on Gemini, Opus in extended thinking on Claude) — it already reasons on its own: no need to tell it to think step by step; instead give it a clear goal and criteria.

🎚️ The reflex: before writing, spot which mode you’re talking to. Simple/fast task → fast mode (+ CoT if needed). Complex task (analysis, code, strategy) → reasoning mode, goal-oriented prompt.

Hands-on case: the same prompt on all three. Take a structured prompt with our 4 ingredients:

[ROLE]     You are an experienced customer-service manager.
[CONTEXT]  Online shop of eco-friendly shoes.
[TASK]     Reply to this 1-star review: "3rd order late, shameful service."
[FORMAT]   4 sentences max, empathetic tone, propose a goodwill gesture.
           If information is missing, don't invent it.

This same prompt, sent to all three, will give all-correct but nuanced answers (illustrative):

• 🟢 ChatGPT — efficient and structured, gets straight to the point, quickly proposes a promo code.
• 🟣 Claude — warmer and more nuanced phrasing, careful with empathy and fluidity.
• 🔵 Gemini — concise, and can chain into the Google ecosystem (“would you like me to send it via Gmail?”).

Caution

⚠️ To be read honestly: these differences are subtle and shifting — from one month to the next, the “best” on this task can change. What doesn’t change: a prompt that’s well structured gets the best out of each. A vague prompt will give three mediocre answers, no matter the assistant.

🧭 How to choose (express guide)

• ✍️ Write / refine wording → Claude in the lead, ChatGPT right behind.
• 🧮 Reason / code → the “reasoning” modes (Claude Opus, ChatGPT Thinking/Pro, Gemini Pro).
• 🖼️ Multimodal (images, voice, screenshots) → ChatGPT or Gemini.
• 📚 Very long document / very large context → Gemini (and the other “large-context” options).
• 🇪🇺 European / open-source / tight-budget option → Mistral, Llama, DeepSeek (see Level 1).

🧪 Mini-exercise (5 min): take the customer-service prompt above, send it as is to two assistants. Note: which one has the tone you would have chosen? Which one invented a piece of information? You’ve just done a home-made benchmark — the only reliable way to choose.

⏳ Freshness note: model names and rankings change every 2-3 months. By the time you read this, the versions above may have already evolved. The reasoning stays valid: portable prompt, choose by task, test it yourself.

✅ In short

• A good prompt is portable: no need to rewrite it from one assistant to another.
• No assistant is “the best” in absolute terms: each shines on certain tasks.
• Spot the mode (fast vs reasoning) before writing — it changes how you use CoT.
• The only reliable judge: test 2 candidates on YOUR task (home-made benchmark).